Pith. sign in

REVIEW 4 major objections 7 minor 73 references

An Active Contour Model for Silhouette Vectorization using B\'ezier Curves

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Minimizing a geodesic active-contour energy over cubic Bézier curves reduces silhouette vectorization error by 15 to 54 percent relative to Inkscape, Adobe Illustrator, and a curvature-based baseline on three test silhouettes.

desk verdict Genuinely new Bezier active-contour formulation with an honest but much too thin experimental section; worth refereeing, not worth accepting as is. read the letter →

arxiv 2505.05132 v2 pith:X6QJYOFB submitted 2025-05-08 cs.GR cs.CVmath.FA

Luis Alvarez , Jean-Michel Morel This is my paper · ORCID
classification cs.GRcs.CVmath.FA MSC 65D1768U1068U05
keywords silhouettevectorizationcubicBéziercurvesactivecontourmodelgeodesicimagetracingcurvefittingvariationalmethodsvectorgraphics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes a post-processing step for silhouette vectorization: take any initial set of cubic Bézier curves approximating a shape boundary, and refine it by minimizing a geodesic active-contour energy that measures how far each point of the Bézier curves lies from the true silhouette. The minimization moves the curves' end points along the boundary, rotates the tangent directions at regular points, and re-estimates the Bézier control parameters. On three test silhouettes the refinement lowers the average distance between the boundary and the vectorization produced by Inkscape, Adobe Illustrator, and a curvature-based baseline by 15 to 54 percent. The interest is that the model is initialization-agnostic: any vectorization method can feed it, and its output should be closer to the original shape.

What carries the argument

The load-bearing mechanism is the geodesic active contour energy (6), a weighted curve length in which the weight at each point is its Euclidean distance to the silhouette boundary plus an optional regularization constant $w_n$. The curves are cubic Bézier segments parameterized so that each endpoint's tangent is expressed in the local frame $T(\alpha_n), T(\alpha_n)^\perp$; this makes the regular-point tangent constraint a simple zero condition on the perpendicular components. Because the energy decouples into separate integrals when $t_n$ and $\alpha_n$ are fixed, the optimization is an alternating scheme: local searches over $t_n$ and $\alpha_n$, with per-interval gradient descent for the four Bézier parameters initialized by the linear least-squares estimate of equation (8), and the distance to the boundary evaluated through an efficient distance computation.

What would settle it

Run the active contour model on a diverse set of silhouettes using deliberately poor initial guesses and measure both $d(B,C)$ and $d(C,B)$; if for some reasonable initial guess the minimization stalls in a local minimum that leaves large uncovered boundary stretches, or if the average distance increases relative to the initial vectorization, the claim that the method significantly improves any vectorization would fail.

Watch

Extended reading notes

Core claim

The central claim is that minimizing the energy functional $$E(\{t_n,\alpha_n,\lambda_n,\gamma_n,\beta_n,\delta_n\}) = \sum_{n=1}^N \$int_0^{{L_n}}$ (d_C(B_n(s)) + w_n)\,\|B_n'(s)\|\,ds$$ over a collection of cubic Bézier curves drives those curves onto the silhouette boundary $C(t)$. Here $d_C(\bar x)$ is the Euclidean distance from a point $\bar x$ to the boundary, and the parameters include the endpoint positions $t_n$, their tangent orientation angles $\alpha_n$ at regular points, and the Bézier control parameters $\lambda_n,\gamma_n,\beta_n,\delta_n$. By writing the control points in terms of the tangent vectors $T(\alpha_n)$ and its perpendicular, regular points are forced to meet the prescribed tangent by simply setting $\gamma_n=\delta_{n-1}=0$. The paper reports that alternating minimization of this energy, updating each $t_n$ within a small window and each $\alpha_n$ within $\pm 4$ degrees while solving per-interval gradient descents, reduces the average boundary-to-curve distance by 15.46 to 54.10 percent across the three silhouettes and three initializers.

Load-bearing premise

The argument assumes that the alternating minimization of the non-convex energy (6) reliably reaches a useful local minimum from whatever initial vectorization is supplied, so that the reported 15 to 54 percent improvement carries over to silhouettes and initial guesses beyond the three tested.

Editorial extensions

If this is right

  • Any vectorization method can be used as the initializer; in the reported experiments, Inkscape, Adobe Illustrator, and the curvature baseline all improved after the active contour refinement.
  • Both one-sided distances $d(B,C)$ and $d(C,B)$ decrease in the reported experiments, so the refined curves both lie close to the boundary and adequately cover it.
  • Raising $w_n$ on a particular Bézier segment shortens that segment, giving a built-in regularity control for removing undesirable irregularities.
  • Per-interval decoupling of the parameter estimation keeps the optimization cost roughly linear in the number of Bézier segments.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the energy is non-convex, the reported gains likely depend on the starting vectorization; in broader use, feeding the model several initial guesses and keeping the lowest-energy output would probably improve robustness.
  • The energy is asymmetric, measuring only the distance from the Bézier curves to the boundary; the reported $d(C,B)$ improvements suggest the asymmetry is mild on these shapes, but a symmetric variant would be safer for shapes with long thin protrusions.
  • The method should extend to open curves and to multiple connected components by treating each Jordan curve independently, as the paper notes; a natural test is whether the gains persist on silhouettes with fine detail or noise.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes a variational active contour model that refits a collection of cubic Bézier curves to a closed silhouette boundary C(t). The energy in Eq. (6) is a geodesic-active-contour integral of the distance to C, weighted by curve length; the unknowns are the curve endpoints t_n, tangent orientations α_n at regular points, and the per-interval Bézier parameters λ_n, γ_n, β_n, δ_n. Minimization is performed by alternating local updates: for fixed endpoints and orientations, the Bézier parameters are obtained by gradient descent on the per-interval functional (7), initialized by the linear least-squares estimate (8); for each n, t_n and α_n are then searched over small windows. Experiments on three silhouettes report reductions of both d(B,C) and d(C,B) for Inkscape, Adobe Illustrator, and a curvature-based baseline, and the paper demonstrates an optional length-regularization parameter w_n.

Significance. The active-contour formulation is clean and well motivated: the tangent-vector parameterization in Eq. (4) makes regularity constraints simple, and the per-interval decoupling for fixed (t_n, α_n) is a practical algorithmic contribution. The paper is also honest about the metric asymmetry, reporting d(C,B), which is not the minimized quantity, and explicitly discussing the symmetric alternative (12). If the reported improvements were confirmed on a broader benchmark with neutral baselines and a robustness analysis, the method would be a useful post-processing tool for silhouette tracing. The main weakness is that the central claim of significant improvement currently rests on three example images, a single run per method, and a heuristic optimization whose convergence is not analyzed.

major comments (4)
  1. [§4, Eqs. (7)–(9)] The alternating minimization is heuristic: each t_n is updated only within [t_n−2, t_n+2], each α_n within ±4 degrees, the inner Bézier-parameter problem (7) is solved by gradient descent from the linear estimate (8), and the outer iteration stops when the energy is no longer 'significantly reduced.' No convergence argument, monotonicity guarantee, or sensitivity study for the window sizes is provided. Since energy (6) is non-convex, the claimed 15–54% improvements are asserted to hold from arbitrary initial vectorizations without evidence that the coordinate-descent scheme reaches a useful local minimum beyond the three reported runs.
  2. [§5, Table 1] The experimental support is thin. Only three silhouettes are tested, each with one run per method; no error bars, no multiple initializations, and no significance tests are reported. With nine numerical comparisons (three methods × three silhouettes), the variability is unknown, so the claim of a 'significant reduction' is not statistically established. Additional experiments on more shapes, with several restarts or perturbations, would be needed to support the generality of the improvement.
  3. [Appendix, MaxDist parameter; Table 1] The curvature baseline appears calibrated to be coarse: the Appendix states that MaxDist = 6 'provides a few interpolation values allowing a better illustration of the performance.' This makes the comparison with the proposed method favorable by construction. A fair evaluation should sweep MaxDist or otherwise select the baseline at a comparable accuracy/complexity trade-off, and should report results at matched node counts or under a common complexity budget, since the number of nodes differs across methods and the paper explicitly declines to consider it.
  4. [Eq. (6), Eq. (10), Table 1] The primary metric d(B,C) in Eq. (10) has the same numerator as the energy (6) being minimized, so a reduction in d(B,C) is partly built into the optimization. The paper correctly acknowledges this and reports d(C,B), which is not minimized; those d(C,B) improvements provide independent evidence in principle, but they come from the same three cases and the same single runs, so the concern is only partially resolved. A larger experiment set or a comparison on the symmetric energy (12) would strengthen the case.
minor comments (7)
  1. [§3, Eq. (6)] The display of the energy (6) is corrupted: the summation appears as 'NX n=1' with a broken N; please fix the typesetting.
  2. [§5, Table 1] The column header 'Var. Perc.' should be defined explicitly (presumably 'relative variation in percent'), and the units of the distance values should be stated (likely pixels, given the 1024×1024 images).
  3. [§4, Eq. (9)] The statement that (t_min, α_min) is the argmin over [t_n−2, t_n+2] × [α_n−r_α, α_n+r_α] is ambiguous because the per-interval parameters are re-optimized for each candidate pair; please clarify whether the search is performed on a discretized grid or on the continuous range using a generic optimizer.
  4. [§4, step 2] The stopping criterion 'significantly reduced' should be made quantitative, for example by specifying a relative energy decrease threshold or a maximum number of iterations.
  5. [References] References [45] and [63] cite the same article (Caselles, Kimmel, and Sapiro, 'Geodesic Active Contours') with different identifiers; please merge them.
  6. [General] No code or data repository is provided, which limits reproducibility of the quantitative claims; making the three silhouette images and the parameter settings publicly available would be helpful.
  7. [Fig. 5] The regularization effect in Fig. 5 is shown only for one zoomed section; a quantitative comparison of distances before and after applying w_n > 0 would be more informative.

Circularity Check

1 steps flagged · score 6.0 of 10

The reported d(B,C) improvement is the optimized energy itself; independent d(C,B) evidence prevents full circularity.

  1. self definitional [Section 3, Eq. (6); Section 5, Eq. (10) and Table 1]
    "In our experiments, we always initially fix wn≡ 0, which in general provides good results. ... The distance d(B,C ) is closely related to the proposed active contour model because our method aims at optimizing the B´ ezier curves by minimizing the term in the numerator of (10)."

    With the default wn=0, Eq. (6) is E = Σ_n ∫_0^{L_n} d_C(B_n(s)) ||B'_n(s)|| ds, which is exactly the numerator of the reported metric d(B,C) in Eq. (10). The alternating minimization in Section 4 is a descent on this same E, and its stopping rule is 'if the functional (6) is significantly reduced, we repeat; otherwise, we stop'. Therefore the Table 1 reductions in d(B,C) are the objective's own decrease, not an independent test of approximation quality. The paper is transparent about this. The independent evidence is the d(C,B) column, which is not minimized; because the paper reports both metrics, the circularity is partial rather than total.

full rationale

The only genuine circular component is the d(B,C) metric: its numerator coincides with the energy functional (6) being minimized, so improvement in that column is largely by construction. The paper honestly acknowledges this and also reports d(C,B), a distance that is not minimized and which also improves in the three experiments; that provides non-circular support for the method. No load-bearing self-citation chain was found: references to the authors' prior work concern standard linear-system derivations ([31]), a distance-transform implementation ([69]), and prior use of the same geodesic weight ([66]), none of which substitutes for the present optimization. The remaining concerns (no convergence guarantees, only three silhouettes, no error bars, and the curvature baseline's MaxDist=6 chosen for illustration) are correctness/robustness issues, not circularity. Overall, the central claim rests partly on a by-construction metric reduction and partly on independent evidence, warranting a partial-circularity score of 6.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim depends on several hand-chosen parameters, chiefly the curvature baseline's MaxDist, which is explicitly tuned to render the illustration favorable. The optimization's convergence is heuristic. The mathematical ingredients (distance transform, cubic Hermite interpolation) are standard. No new entities are introduced.

free parameters (7)
  • MaxDist (curvature baseline) = 6
    Threshold for adding regular points in the curvature method; chosen because it 'provides a few interpolation values allowing a better illustration of the performance'.
  • MinLength = 25
    Minimum arc length between selected points in the curvature baseline.
  • sigma (Gaussian scale for tangent/cornerness) = 20
    Scale parameter for the cornerness measure and tangent estimation in the curvature baseline.
  • kappa_min = 0.5
    Cornerness threshold for selecting corner points in the curvature baseline.
  • Search window for t_n = +/-2
    Local search range for endpoint positions in the coordinate descent; hand-chosen.
  • Search range for alpha_n = +/-4 degrees
    Local search range for tangent orientations; hand-chosen.
  • Regularization weight w_n = 0 (default), 30 (one example)
    Optional length penalty; set to 0 in all main experiments and to 30 for a regularization illustration.
assumptions (5)
  • standard math The input boundary C is a piecewise-smooth closed curve and its distance function d_C is computable and differentiable almost everywhere.
    Used to define the energy (6) and to run gradient descent; standard in geometric active contour models.
  • domain assumption Silhouettes are clean binary shapes given by a single Jordan curve.
    All experiments use three synthetic-style clean silhouettes; real noisy or multi-component shapes are not tested (Section 5).
  • domain assumption Inkscape and Illustrator SVG outputs contain only corner points, so no tangent constraints are imposed on their initial traces.
    The authors state they 'assume that all points have two half-tangents and, therefore, are corners' (Section 5).
  • ad hoc to paper Coordinate descent with the stated local search windows converges to a local minimum that improves the vectorization.
    No convergence proof is provided; the reported improvements depend on this empirical behavior (Section 4).
  • domain assumption Cubic Bézier curves can faithfully represent the boundary with the chosen corner/regular segmentation.
    The representation (4) assumes the boundary is well approximated by Hermite-style cubic segments with endpoint tangents; failure would show as systematic bias in the fit.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Active Contour Model for Silhouette Vectorization using B\'ezier Curves." pith.science (2026). https://pith.science/paper/X6QJYOFB

@misc{pith2026250505132,
  author       = {Pith},
  title        = {Pith review of: An Active Contour Model for Silhouette Vectorization using B\'ezier Curves},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X6QJYOFB}},
  note         = {Machine review of arXiv:2505.05132}
}
read the original abstract

In this paper, we propose an active contour model for silhouette vectorization using cubic B\'ezier curves. Among the end points of the B\'ezier curves, we distinguish between corner and regular points where the orientation of the tangent vector is prescribed. By minimizing the distance of the B\'ezier curves to the silhouette boundary, the active contour model optimizes the location of the B\'ezier curves end points, the orientation of the tangent vectors in the regular points, and the estimation of the B\'ezier curve parameters. This active contour model can use the silhouette vectorization obtained by any method as an initial guess. The proposed method significantly reduces the average distance between the silhouette boundary and its vectorization obtained by the world-class graphic software Inkscape, Adobe Illustrator, and a curvature-based vectorization method, which we introduce for comparison. Our method also allows us to impose additional regularity on the B\'ezier curves by reducing their lengths.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 72 canonical work pages

  1. [1]

    Academic Press, (2016)

    Bachmann, T.: Perception of Pixelated Images. Academic Press, (2016)

    work page 2016

  2. [2]

    Journal of experimental Psychology 53(4), 221 (1957)

    Attneave, F.: Physical determinants of the judged complexity of shapes. Journal of experimental Psychology 53(4), 221 (1957)

    work page 1957

  3. [3]

    Elder, J.H.: Are edges incomplete? Interna- tional Journal of Computer Vision 34(2-3), 97–122 (1999)

    work page 1999

  4. [4]

    Torralba, A.: How many pixels make an image? Visual neuroscience 26(1), 123–131 (2009)

    work page 2009

  5. [5]

    IEEE Transactions on Visual- ization and Computer Graphics (2022)

    Tian, X., G¨ unther, T.: A survey of smooth vector graphics: Recent advances in repre- sentation, creation, rasterization and image vectorization. IEEE Transactions on Visual- ization and Computer Graphics (2022)

    work page 2022

  6. [6]

    arXiv preprint arXiv:2306.06441 (2023)

    Dziuba, M., Jarsky, I., Efimova, V., Filchenkov, A.: Image vectorization: a review. arXiv preprint arXiv:2306.06441 (2023)

    arXiv 2023

  7. [7]

    Image Processing On Line (2021)

    He, Y., Kang, S.H., Alvarez, L.: Finding the skeleton of 2d shape and contours: Imple- mentation of hamilton-jacobi skeleton. Image Processing On Line (2021)

    work page 2021

  8. [8]

    Journal of Mathematical Imaging and Vision, 1–16 (2022)

    He, Y., Kang, S.H., Morel, J.-M.: Silhouette vectorization by affine scale-space. Journal of Mathematical Imaging and Vision, 1–16 (2022)

    work page 2022

Show all 73 references
  1. [9]

    Journal of Mathematical Imaging and Vision 65(6), 874–893 (2023)

    He, Y., Kang, S.H., Morel, J.-M.: Topology- and perception-aware image vectorization. Journal of Mathematical Imaging and Vision 65(6), 874–893 (2023)

    2023

  2. [10]

    ACM Transactions on Graph- ics (TOG) 27(3), 1–8 (2008)

    Orzan, A., Bousseau, A., Winnem¨ oller, H., Barla, P., Thollot, J., Salesin, D.: Diffusion curves: a vector representation for smooth- shaded images. ACM Transactions on Graph- ics (TOG) 27(3), 1–8 (2008)

    2008

  3. [11]

    IEEE Trans- actions on Pattern Analysis and Machine Intelligence 28(6), 890–904 (2006)

    Hilaire, X., Tombre, K.: Robust and accurate vectorization of line drawings. IEEE Trans- actions on Pattern Analysis and Machine Intelligence 28(6), 890–904 (2006)

    2006

  4. [12]

    ACM Transac- tions on Graphics (TOG) 32(1), 1–11 (2013)

    Noris, G., Hornung, A., Sumner, R.W., Sim- mons, M., Gross, M.: Topology-driven vector- ization of clean line drawings. ACM Transac- tions on Graphics (TOG) 32(1), 1–11 (2013)

    2013

  5. [13]

    simplicity: a global approach to line drawing vectorization

    Favreau, J.-D., Lafarge, F., Bousseau, A.: Fidelity vs. simplicity: a global approach to line drawing vectorization. ACM Transac- tions on Graphics (TOG) 35(4), 1–10 (2016)

    2016

  6. [14]

    IEEE Transactions on Visualiza- tion and Computer Graphics 15(4), 618–629 (2009)

    Zhang, S.-H., Chen, T., Zhang, Y.-F., Hu, S.-M., Martin, R.R.: Vectorizing cartoon ani- mations. IEEE Transactions on Visualiza- tion and Computer Graphics 15(4), 618–629 (2009)

    2009

  7. [15]

    IEEE transactions on visualiza- tion and computer graphics 22(2), 1063–1075 (2015)

    Yang, M., Chao, H., Zhang, C., Guo, J., Yuan, L., Sun, J.: Effective clipart image vectorization through direct optimization of bezigons. IEEE transactions on visualiza- tion and computer graphics 22(2), 1063–1075 (2015)

    2015

  8. [16]

    ACM Transactions on Graphics (TOG) 36(6), 1–11 11 (2017)

    Favreau, J.-D., Lafarge, F., Bousseau, A.: Photo2clipart: Image abstraction and vector- ization using layered linear gradients. ACM Transactions on Graphics (TOG) 36(6), 1–11 11 (2017)

    2017

  9. [17]

    Image Processing On Line 13, 22–37 (2023)

    He, Y., Kang, S.H., Morel, J.-M.: Binary shape vectorization by affine scale-space. Image Processing On Line 13, 22–37 (2023)

    2023

  10. [18]

    In: 2023 IEEE Inter- national Conference on Image Processing (ICIP), pp

    He, Y., Kang, S.H., Morel, J.-M.: Viva: a variational image vectorization algorithm on dual-primal graph pairs. In: 2023 IEEE Inter- national Conference on Image Processing (ICIP), pp. 1285–1289 (2023). IEEE

    2023

  11. [19]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Reddy, P., Gharbi, M., Lukac, M., Mitra, N.J.: Im2vec: Synthesizing vector graphics without vector supervision. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7342– 7351 (2021)

    2021

  12. [20]

    ACM Transactions on Graphics (TOG) 39(6), 1–15 (2020)

    Li, T.-M., Luk´ aˇ c, M., Gharbi, M., Ragan- Kelley, J.: Differentiable vector graphics ras- terization for editing and learning. ACM Transactions on Graphics (TOG) 39(6), 1–15 (2020)

    2020

  13. [21]

    iuniverse Bloomington, (2000)

    Ferraiolo, J., Jun, F., Jackson, D.: Scal- able Vector Graphics (SVG) 1.0 Specification. iuniverse Bloomington, (2000)

    2000

  14. [22]

    https://www

    Adobe: Adobe Illustrator. https://www. adobe.com/products/illustrator.html

  15. [23]

    ACM Transactions on Graphics (TOG) 26(3), 11 (2007)

    Sun, J., Liang, L., Wen, F., Shum, H.-Y.: Image vectorization using optimized gradi- ent meshes. ACM Transactions on Graphics (TOG) 26(3), 11 (2007)

    2007

  16. [24]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Lopes, R.G., Ha, D., Eck, D., Shlens, J.: A learned representation for scalable vector graphics. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 7930–7939 (2019)

    2019

  17. [25]

    ACM Transactions on Graphics (TOG) 41(4), 1–11 (2022)

    Vinker, Y., Pajouheshgar, E., Bo, J.Y., Bachmann, R.C., Bermano, A.H., Cohen- Or, D., Zamir, A., Shamir, A.: Clipasso: Semantically-aware object sketching. ACM Transactions on Graphics (TOG) 41(4), 1–11 (2022)

    2022

  18. [26]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Ma, X., Zhou, Y., Xu, X., Sun, B., Filev, V., Orlov, N., Fu, Y., Shi, H.: Towards layer-wise image vectorization. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 16314–16323 (2022)

    2022

  19. [27]

    PhD the- sis, Monterey, California

    Pastva, T.A.: Bezier curve fitting. PhD the- sis, Monterey, California. Naval Postgraduate School (1998)

    1998

  20. [28]

    Selinger, P.: Potrace : a polygon- based tracing algorithm. (2003). https://api.semanticscholar.org/CorpusID:1419652

    2003

  21. [29]

    Pattern Recognition Letters 19(9), 821–828 (1998) https: //doi.org/10.1016/S0167-8655(98)00069-5

    Cinque, L., Levialdi, S., Malizia, A.: Shape description using cubic polyno- mial bezier curves. Pattern Recognition Letters 19(9), 821–828 (1998) https: //doi.org/10.1016/S0167-8655(98)00069-5

    1998 doi

  22. [30]

    In: 10th Annual Conference on Computer Graphics and Inter- active Techniques, pp

    Plass, M., Stone, M.: Curve-fitting with piecewise parametric cubics. In: 10th Annual Conference on Computer Graphics and Inter- active Techniques, pp. 229–239 (1983)

    1983

  23. [31]

    Journal of Mathematical Imaging and Vision 64, 41–56 (2022)

    He, Y., Kang, S.H., Morel, J.-M.: Silhouette vectorization by affine scale-space. Journal of Mathematical Imaging and Vision 64, 41–56 (2022)

    2022

  24. [32]

    Journal of Mathematical Imaging and Vision 1, 1–20 (2023)

    He, Y., Kang, S.H., Morel, J.M.: Topology- and perception-aware image vectorization. Journal of Mathematical Imaging and Vision 1, 1–20 (2023)

    2023

  25. [33]

    Computer Graphics and Image Processing 1(4), 360–372 (1972)

    Pavlidis, T.: Segmentation of pictures and maps through functional approximation. Computer Graphics and Image Processing 1(4), 360–372 (1972)

    1972

  26. [34]

    Communi- cations on pure and applied mathematics (1989)

    Mumford, D.B., Shah, J.: Optimal approxi- mations by piecewise smooth functions and associated variational problems. Communi- cations on pure and applied mathematics (1989)

    1989

  27. [35]

    Morel, J.-M., Solimini, S.: Variational Meth- ods in Image Segmentation: with Seven Image Processing Experiments vol. 14. Springer, (2012) 12

    2012

  28. [36]

    Applied mathematical sciences 147 (2006)

    Kornprobst, G.A.P., Aubert, G.: Mathemat- ical problems in image processing. Applied mathematical sciences 147 (2006)

    2006

  29. [37]

    SIAM Journal on Imaging Sciences 6(1), 368–390 (2013)

    Cai, X., Chan, R., Zeng, T.: two-stage image seagmentation method using a convex variant of the mumford–shah model and threshold- ing. SIAM Journal on Imaging Sciences 6(1), 368–390 (2013)

    2013

  30. [38]

    In: 2009 IEEE 12th International Conference on Computer Vision, pp

    Pock, T., Cremers, D., Bischof, H., Cham- bolle, A.: An algorithm for minimizing the mumford-shah functional. In: 2009 IEEE 12th International Conference on Computer Vision, pp. 1133–1140 (2009). IEEE

    2009

  31. [39]

    In: Proceedings IEEE Workshop on Variational and Level Set Meth- ods in Computer Vision, pp

    Chan, T.F., Vese, L.A.: A level set algorithm for minimizing the mumford-shah functional in image processing. In: Proceedings IEEE Workshop on Variational and Level Set Meth- ods in Computer Vision, pp. 161–168 (2001). IEEE

    2001

  32. [40]

    Commu- nications on Pure and Applied Mathematics 43(8), 999–1036 (1990)

    Ambrosio, L., Tortorelli, V.M.: Approxima- tion of functional depending on jumps by elliptic functional via t-convergence. Commu- nications on Pure and Applied Mathematics 43(8), 999–1036 (1990)

    1990

  33. [41]

    International journal of computer vision 1(4), 321–331 (1988)

    Kass, M., Witkin, A., Terzopoulos, D.: Snakes: Active contour models. International journal of computer vision 1(4), 321–331 (1988)

    1988

  34. [42]

    IEEE Transactions on image processing 7(3), 359–369 (1998)

    Xu, C., Prince, J.L.: Snakes, shapes, and gradient vector flow. IEEE Transactions on image processing 7(3), 359–369 (1998)

    1998

  35. [43]

    In: [1988 Pro- ceedings] Second International Conference on Computer Vision, pp

    Amini, A.A., Tehrani, S., Weymouth, T.E.: Using dynamic programming for minimiz- ing the energy of active contours in the presence of hard constraints. In: [1988 Pro- ceedings] Second International Conference on Computer Vision, pp. 95–99 (1988). IEEE

    1988

  36. [44]

    Journal of Mathematical Imaging and vision 28(2), 151–167 (2007)

    Bresson, X., Esedo¯ glu, S., Vandergheynst, P., Thiran, J.-P., Osher, S.: Fast global mini- mization of the active contour/snake model. Journal of Mathematical Imaging and vision 28(2), 151–167 (2007)

    2007

  37. [45]

    International jour- nal of computer vision 22, 61–79 (1997)

    Caselles, V., Kimmel, R., Sapiro, G.: Geodesic active contours. International jour- nal of computer vision 22, 61–79 (1997)

    1997

  38. [46]

    Springer, (2003)

    Cao, F.: Geometric Curve Evolution and Image Processing. Springer, (2003)

    2003

  39. [47]

    Atti della Accademia Nazionale dei Lincei

    Miranda, M.: Caccioppoli sets. Atti della Accademia Nazionale dei Lincei. Classe di Scienze Fisiche, Matematiche e Naturali. Rendiconti Lincei. Matematica e Applicazioni 14(3), 173–177 (2003)

    2003

  40. [48]

    Journal of the European Mathe- matical Society 3(1), 39–92 (2001)

    Ambrosio, L., Morel, J.-M., Masnou, S., Caselles, V.: Connected components of sets of finite perimeter and applications to image processing. Journal of the European Mathe- matical Society 3(1), 39–92 (2001)

    2001

  41. [49]

    Jour- nal of Differential geometry 26(2), 285–314 (1987)

    Grayson, M.A.: The heat equation shrinks embedded plane curves to round points. Jour- nal of Differential geometry 26(2), 285–314 (1987)

    1987

  42. [50]

    Chapman and Hall/CRC, (2001)

    Chou, K.-S., Zhu, X.-P.: The Curve Short- ening Problem. Chapman and Hall/CRC, (2001)

    2001

  43. [51]

    IEEE transactions on pattern analysis and machine intelligence (1), 2–14 (1986)

    Asada, H., Brady, M.: The curvature primal sketch. IEEE transactions on pattern analysis and machine intelligence (1), 2–14 (1986)

    1986

  44. [52]

    Courier Dover Publications, (2016)

    Do Carmo, M.P.: Differential Geometry of Curves and Surfaces: Revised and Updated Second Edition. Courier Dover Publications, (2016)

    2016

  45. [53]

    Jour- nal of computational physics 79(1), 12–49 (1988)

    Osher, S., Sethian, J.A.: Fronts propagating with curvature-dependent speed: Algorithms based on hamilton-jacobi formulations. Jour- nal of computational physics 79(1), 12–49 (1988)

    1988

  46. [54]

    International journal of com- puter vision 11(1), 25–44 (1993)

    Sapiro, G., Tannenbaum, A.: Affine invari- ant scale-space. International journal of com- puter vision 11(1), 25–44 (1993)

    1993

  47. [55]

    SIAM Journal on Applied Mathematics 57(1), 176–194 (1997)

    Tannenbaum, A., Sapiro, G., Olver, P.J.: Invariant geometric evolutions of surfaces and volumetric smoothing. SIAM Journal on Applied Mathematics 57(1), 176–194 (1997)

    1997

  48. [56]

    Journal of the American Mathemati- cal Society 11(3), 601–634 (1998)

    Angenent, S., Sapiro, G., Tannenbaum, A.: 13 On the affine heat equation for non-convex curves. Journal of the American Mathemati- cal Society 11(3), 601–634 (1998)

    1998

  49. [57]

    IEEE Transactions on Image Processing 7(3), 411–420 (1998)

    Moisan, L.: Affine plane curve evolution: A fully consistent scheme. IEEE Transactions on Image Processing 7(3), 411–420 (1998)

    1998

  50. [58]

    In: 2021 IEEE International Confer- ence on Image Processing (ICIP), pp

    He, Y., Kang, S.H., Morel, J.-M.: Accu- rate silhouette vectorization by affine scale- space. In: 2021 IEEE International Confer- ence on Image Processing (ICIP), pp. 1539– 1543 (2021). IEEE

    2021

  51. [59]

    In: 2022 IEEE International Conference on Image Process- ing (ICIP), pp

    He, Y., Kang, S.H., Morel, J.-M.: Vector- izing images of any size. In: 2022 IEEE International Conference on Image Process- ing (ICIP), pp. 816–820 (2022). IEEE

    2022

  52. [60]

    International Journal of Computer Vision 25, 95–107 (1997)

    Alvarez, L., Morales, F.: Affine morphologi- cal multiscale analysis of corners and multiple junctions. International Journal of Computer Vision 25, 95–107 (1997)

    1997

  53. [61]

    In: International Conference on Scale Space and Variational Methods in Computer Vision, pp

    Alvarez, L.: Corner detection using the affine morphological scale space. In: International Conference on Scale Space and Variational Methods in Computer Vision, pp. 29–40 (2017). Springer

    2017

  54. [62]

    arXiv preprint arXiv:1301.3352 (2013)

    Magni, A., Mantegazza, C., Novaga, M.: Motion by curvature of planar networks ii. arXiv preprint arXiv:1301.3352 (2013)

    2013 arXiv

  55. [63]

    International Journal of Computer Vision22, 61–79 (1997)

    Caselles, V., Kimmel, R., Sapiro, G.: Geodesic Active Contours. International Journal of Computer Vision22, 61–79 (1997)

    1997

  56. [64]

    International Journal of Computer Vision 50, 315–328 (2002)

    Chen, Y., Tagare, H.D., Thiruvenkadam, S., Huang, F., Wilson, D., Gopinath, K.S., Briggs, R.W., Geiser, E.A.: Using prior shapes in geometric active contours in a vari- ational framework. International Journal of Computer Vision 50, 315–328 (2002)

    2002

  57. [65]

    In: Duncan, J.S., Gerig, G

    Liu, H., Chen, Y., Ho, H.P., Shi, P.: Geodesic active contours with adaptive neighboring influence. In: Duncan, J.S., Gerig, G. (eds.) Medical Image Computing and Computer- Assisted Intervention – MICCAI 2005, pp. 741–748. Springer, Berlin, Heidelberg (2005)

    2005

  58. [66]

    Alvarez, L.: 3D curve regularization. Rev. Real Acad. Cienc. Exactas Fis. Nat. Ser. A-Mat 116 (106) , 1–10 (2022)

    2022

  59. [67]

    Journal of Real-Time Image Processing 14, 793–802 (2018)

    Cuenca, C., Gonz´ alez, E., Trujillo, A., Esclarin, J., Mazorra, L., Alvarez, L., Mar- tinez, J.A., Tahoces, P.G., Carreira, J.M.: Fast and accurate circle tracking using active contour models. Journal of Real-Time Image Processing 14, 793–802 (2018)

    2018

  60. [68]

    Journal of Mathematical Imaging and Vision 60 (7) , 1–10 (2018)

    Alvarez, L., Gonz´ alez, E., Cuenca, C., Tru- jillo, A., Tahoces, P.G., Carreira, J.M.: Ellipse motion estimation using parametric snakes. Journal of Mathematical Imaging and Vision 60 (7) , 1–10 (2018)

    2018

  61. [69]

    IPOL Journal : Image Processing On Line11, 18–36 (2021) Appendix

    He, Y., Kang, S.H., Alvarez, L.: Finding the skeleton of 2d shape and contours: Imple- mentation of hamilton-jacobi skeleton. IPOL Journal : Image Processing On Line11, 18–36 (2021) Appendix. A basic vectorization algorithm based on the curvature of the silhouette contour We d...

    2021

  62. [70]

    This is the most important parame- ter of the algorithm and determines the number of end points of the B´ ezier curves

    MaxDist : the maximum distance allowed between the B´ ezier curves and the original curveC(t). This is the most important parame- ter of the algorithm and determines the number of end points of the B´ ezier curves. In all the experiments presented, we fix MaxDist = 6 which pro...

  63. [71]

    In all the experiments presented, we fix MinLength = 25

    MinLength : the minimum length distance (along the curve C(t)) between the selected points. In all the experiments presented, we fix MinLength = 25

  64. [72]

    In all the experiments presented, we fixσ = 20

    σ : scale parameter used to estimate the corner- ness measure and the tangent vector to C(t). In all the experiments presented, we fixσ = 20

  65. [73]

    In all the experiments presented, we fix κmin = 0.5

    κmin : the threshold value of the cornerness measure to define the set Corners. In all the experiments presented, we fix κmin = 0.5. 15

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.